Bump CE and fs2 to the latest version

[aartigao]

Fixes #1234.

NOTE: I had to bump Scala 3 lib because since 3.8.0 fs2 uses 3.3.x

[aartigao]

In addition to the changes suggested, while removing the TODO in produceRecord[F[_], K, V] method, I noticed the usage of Promise there inside Callback.

Consider the following code that uses directly the Producer API method with signature Future<RecordMetadata> send(ProducerRecord<K, V> record); and waits asynchronously to the result of the returned Future using F.async_ 👇🏽

private[kafka] def produceRecord[F[_], K, V](
  keySerializer: KeySerializer[F, K],
  valueSerializer: ValueSerializer[F, V],
  producer: KafkaByteProducer,
  blocking: Blocking[F]
)(
  implicit F: Async[F]
): ProducerRecord[K, V] => F[F[(ProducerRecord[K, V], RecordMetadata)]] =
  record =>
    asJavaRecord(keySerializer, valueSerializer, record).flatMap { javaRecord =>
      blocking {
        producer.send(javaRecord)
      }.map { future =>
        F.async_ { cb =>
          Try(future.get) match {
            case Success(metadata)               => cb(Right((record, metadata)))
            case Failure(ex: ExecutionException) => cb(Left(ex.getCause))
            case Failure(exception)              => cb(Left(exception))
          }
        }
      }
    }

According to async_ documentation:

This function can be thought of as a safer, lexically-constrained version of `Promise`,
where `IO` is like a safer, lazy version of `Future`.

What do you think about it? TBH I'm not really sure if there's any benefit on using async_ or if it's actually worse 🤷🏽 (the Try(future.get) part is the one that concerns me).

cc @vlovgr

[vlovgr]

What do you think about it? TBH I'm not really sure if there's any benefit on using async_ or if it's actually worse 🤷🏽 (the Try(future.get) part is the one that concerns me).

@aartigao With the new cancellation semantics in cats-effect 3.5.0, F.async_ is uncancelable. Try(future.get) is also blocking, so I'd suggest we keep what you already have in this pull request. 👍